Classification of synthetic aperture radar images using Markov Random Field and textural features

Synthetic aperture radar (SAR) is an active imagery system which allows day-and-night and all-weather acquisitions. SAR images are usually affected by a multiplicative noise depending on the ground reflectivity due to the coherence of the radar wavelength [1]. For this reason, classification of SAR images is not a straightforward task, and pixel based classification algorithms will struggle to achieve decent results. a possible solution to this problem is utilizing the spatial relationship between neighboring pixels. The Statistical dependency between neighboring pixels is modeled by Markov Random Field (MRF). In this paper, we present a novel classification algorithm for SAR images using MRF model. The method is based on an iterative expectation maximization (EM) procedure. The iterative process can be initialized by a texture grade images. In this way we avoid all manual intervention. In addition, we suggest an improvement for an existing classification algorithm [2] by using our EM procedure and texture images for expanding the MRF model to a 3-D model. The algorithm estimates the probability density function (PDF) of each class by a pre-defined, dictionary based, stochastic expectation maximization (SEM) procedure [3].